Document Contents

Output is presented for the following sections:

  1. Calibration Site Maps for User-Selected Attributes
  2. Model Estimation Performance Diagnostics
  3. Model Simulation Performance Diagnostics
  4. Maps of Model Residuals and Observed to Predicted Ratios for the Calibration Sites

Model Estimation Performance Diagnostics

Diagnostics are based on the use of conditioned (monitoring-adjusted) predictions. These predictions provide the most accurate reach predictions for use in calibrating the model. The associated residuals and observed to predicted ratios shown in the following section provide the most relevant measures of the accuracy of the model fit to observed loads.

The diagnostic plots include:

  1. Four-plot panel for observed vs. predicted for loads and yields, and log residuals vs. predicted loads and yields

  2. Four-plot panel for boxplots of residuals and observed/predicted ratios, normal quantile plot of standardized residuals, and plot of squared residuals vs. predicted loads

  3. Plot of conditioned prediction loads vs. unconditioned (simulated) prediction loads

  4. Plots of the observed to predicted ratio vs. the area-weighted mean values of the user-selected explanatory variables for the incremental areas between calibration sites (output only if control setting if_corrExplanVars<-‘yes’ selected and a value of 1 entered for ‘parmCorrGroup’ column in the ‘parameters.csv’ file)

  5. Boxplots of the observed to predicted loads vs. the decile classes of the total drainage area for the calibration sites

  6. Boxplots of the observed to predicted loads vs. the contiguous spatial classes specified by users in the ‘classvar’ control setting (e.g., HUC-4)

  7. Boxplots of the observed to predicted loads vs. the deciles of the land-use class variable specified by users in the ‘class_landuse’ control setting, with the land-use classes expressed as a percentage of the incremental drainage area extending from the calibration site to the nearest upstream site locations

  8. Four-plot panels reported separately for each of the contiguous spatial classes specified for the first variable entry for the ‘classvar[1]’ control setting. The panels include: observed vs. predicted loads, observed vs. predicted yields, log residuals vs. predicted loads, and log residuals vs. predicted yields

Observed vs. predicted for loads and yields and log residuals vs. predicted loads and yields

Boxplots of residuals and observed/predicted ratios, normal quantile plot of standardized residuals, and plot of squared residuals vs. predicted loads

Conditioned prediction loads vs. unconditioned (simulated) prediction loads

Observed to predicted ratio vs. the area-weighted mean values of the user-selected explanatory variables for the incremental areas between calibration sites

Output only if control setting if_corrExplanVars<-‘yes’ selected and a value of 1 entered for ‘parmCorrGroup’ column in the ‘parameters.csv’ file.

Boxplots of the observed to predicted loads vs. the decile classes of the total drainage area for the calibration sites

Boxplots of the observed to predicted loads vs. the contiguous spatial classes specified by users in the ‘classvar’ control setting (e.g., HUC-4)

Boxplots of the observed to predicted loads vs. the deciles of the land-use class variable specified by users in the ‘class_landuse’ control setting

The land-use classes expressed as a percentage of the incremental drainage area extending from the calibration site to the nearest upstream site locations

Four-plot panels reported separately for each of the contiguous spatial classes specified for the first variable entry for the ‘classvar[1]’ control setting

The panels include: observed vs. predicted loads, observed vs. predicted yields, log residuals vs. predicted loads, and log residuals vs. predicted yields

Model Simulation Performance Diagnostics

Diagnostics are based on the use of unconditioned predictions (i.e., predictions that are not adjusted for monitoring loads). These predictions (and the associated residuals and observed to predicted ratios shown in the following section) provide the best measure of the predictive skill of the estimated model in simulation mode. The simulated predictions are computed using mean coefficients from the NLLS model estimated with monitoring-adjusted (conditioned) predictions.

The diagnostic plots include:

  1. Four-plot panel for observed vs. predicted for loads and yields, and log residuals vs. predicted loads and yields

  2. Four-plot panel for boxplots of residuals and observed/predicted ratios, normal quantile plot of standardized residuals, and plot of squared residuals vs. predicted loads

  3. Plot of conditioned prediction loads vs. unconditioned (simulated) prediction loads

  4. Plots of the observed to predicted ratio vs. the area-weighted mean values of the user-selected explanatory variables for the incremental areas between calibration sites (output only if control setting if_corrExplanVars<-‘yes’ selected and a value of 1 entered for ‘parmCorrGroup’ column in the ‘parameters.csv’ file)

  5. Boxplots of the observed to predicted loads vs. the decile classes of the total drainage area for the calibration sites

  6. Boxplots of the observed to predicted loads vs. the contiguous spatial classes specified by users in the ‘classvar’ control setting (e.g., HUC-4)

  7. Boxplots of the observed to predicted loads vs. the deciles of the land-use class variable specified by users in the ‘class_landuse’ control setting, with the land-use classes expressed as a percentage of the incremental drainage area extending from the calibration site to the nearest upstream site locations

  8. Four-plot panels reported separately for each of the contiguous spatial classes specified for the first variable entry for the ‘classvar[1]’ control setting. The panels include: observed vs. predicted loads, observed vs. predicted yields, log residuals vs. predicted loads, and log residuals vs. predicted yields

Maps of Model Residuals and Observed to Predicted Ratios for the Calibration Sites

The maps include:

  1. Log residuals, based on monitoring conditioned predictions (i.e., Model Estimation Log Residuals)

  2. Log residuals, based on the unconditioned predictions (i.e., Model Simulation Log Residuals)

  3. Standardized residuals based on the monitoring conditioned predictions

  4. Ratio of observed to predicted loads for the conditioned predictions (i.e., Model Estimation Ratio)

  5. Ratio of observed to predicted load for the unconditioned predictions (i.e., Model Simulation Ratio)